In a two-way telecommunication system, acoustic echo is the result of a signal output by the loudspeaker being captured by the microphone on the same device, typically called the near-end device. The result of this capture is that the user of the other side of the conversation, the far-end user, hears their own voice as an echo, after a telecommunication lag. Echo is provably detrimental to having a natural conversation and approaches to eliminate echo are needed for speakerphone conversations.
Speaker-phone devices sometimes use a full-duplex acoustic echo cancellation (AEC) algorithm to cancel the echoed speech and to minimize the impact conversational quality. The AEC algorithm works best when a purely linear dependency can be established between the signal sent to the loudspeaker and the signal captured by the microphone.
In current-day mobile devices, the drive towards portability and slimmer industrial design has led to smaller volumes being available inside these devices. Acoustic performance suffers from this constraint in available volume. Smaller and thinner loudspeakers need to be used for thin elegant designs. As a result, lower volumes of air can be moved leading to a reduction in sound pressure levels (SPLs), especially at lower frequencies. When original equipment manufacturers (OEMs) design to specification of a small speaker, then the resultant speakers are not able to achieve desired SPLs. Audio system engineers may overdrive the loudspeakers and power amplifiers to get the desired SPL. However, the constraints on loudspeaker size often result in non-linear output.
In general, loudspeaker vibrations behave differently at small and high amplitudes. The dependency on the amplitude is an indication of nonlinearities inherent in the system. Overdriving speakers also causes non-linear distortion by excessive excursion of the loudspeaker diaphragm, a result of soft clipping. This secondary nonlinear effect is the generation of additional spectral components which are not in the exciting stimulus. They are usually integer multiples of the applied fundamentals as harmonics or inter-modulation distortion between multiple fundamentals. The creation of harmonics is also exploited to overdrive small speakers, because harmonics of low frequency fundamentals can psycho-acoustically fill in for the low frequency loss making the lows end of the audio spectrum sound fuller than they are.
One factor that affects non-linear behavior in small speakers is the Magnetic Force factor BI(x). The magnetic force factor is a function of displacement as the field is not uniform. This has more impact in a smaller speaker. Another factor is Compliance Cms(x) or inversely Stiffness of Suspension Kms(x). This is the restoring force of the suspension increases as a function of the distance. Still another factor is the Mechanical Resistance Rms(v) and this factor depends on the velocity of the coil. The Mechanical Resistance increases with increasing velocity in either direction.
The major cause of non-linearities in the echo-path of a speakerphone application is usually the speaker. All of these factors influence the force applied to the diaphragm, which further affects the displacement and velocity of the voice coil. This is a non-linear feedback loop with interactions between these factors. Bl(x) and Kms(x) generate DC components while Rms(v) causes compression of the fundamental at resonance. These affect Total Harmonic Distortion (THD) and Inter Modulation Distortion (IMD) via feedback. The dominant non-linearities in loudspeakers can have significant impact on the output signal.
Overdriving speakers beyond specifications can also cause loudspeakers to fail because excessive excursion can cause the diaphragm to blow and because excess heat in the voice coil of the speaker can also cause its failure. Electronic feedback compensation may be used for speaker protection. Speaker protection technology measures current and voltage across the voice coil to estimate the temperature and excursion and adaptively limit these below maximum levels for highest possible SPLs and better audibility. This increases life of the loudspeaker while extracting maximum performance.
Speaker protection technology use adaptive algorithms to manage the output signal dynamics, in order to constrain excursion, limit temperature and minimize THD while maximizing the output power. The loudspeaker output varies constantly in time and frequency compared to the signal that the AEC uses. Thus, loudspeaker signal output levels can be temporally highly de-correlated from the far-end signal that an AEC algorithm normally uses as the echo reference signal, i.e. it is no longer a time-invariant even in a quasi-stationary sense.
Small speakers are used in mobile audio because of: constraints on dimensions driven by marketing and industrial design requirements; small space for speaker acoustics (including back volume); thin devices for mobility and convenience of handling; economic decisions by manufacturers; and the control of the Bill of Materials and costs in a highly competitive market.
Small speakers are often overdriven because of engineering and marketing decisions. These speakers need to meet target SPL for use case and induce harmonic distortion of low frequency fundamentals to enhance bass perception especially as low frequency is attenuated in smaller speakers.
The consequence of speaker non-linearities and speaker protection algorithms for AEC algorithms is that the linear time-invariant relationship between the echo reference signal and actual loudspeaker output is not maintained. This can result in poor AEC performance.
There are several problems with such non-linear system identification and previous algorithms were not good enough to match the performance of an AEC under linear conditions. The computational complexities of non-linear AEC (NLAEC) algorithms are also significantly more than linear AEC algorithms. Additionally the need to over-drive small speakers for deriving sufficient output for such use cases has also led to the development of time varying speaker protection algorithms, which dynamically adjust the drive power of the speaker depending on real-time performance parameters derived from the speakers. This can result in reduction of non-linearity but leads a speaker output system that is time varying. This also causes problems on convergence of AEC algorithms, which typically work of reference signals derived before the speaker protection algorithm.
All of these problems have resulted in some user dissatisfaction with previous approaches.
Skilled artisans will appreciate that elements in the figures are illustrated for simplicity and clarity. It will further be appreciated that certain actions and/or steps may be described or depicted in a particular order of occurrence while those skilled in the art will understand that such specificity with respect to sequence is not actually required. It will also be understood that the terms and expressions used herein have the ordinary meaning as is accorded to such terms and expressions with respect to their corresponding respective areas of inquiry and study except where specific meanings have otherwise been set forth herein.